1.1.1 Sample Information

A brief sample information is generated from the submission table for the following analysis.

1.1.2 MERSCOPE Data Quality Summary

The summaries present the data quality assessment automatically generated by MERSCOPE for each experiment. We mainly focus on the transcripts level for each sample. So we’re looking for high density in transcripts, based on the transcripts count per field of view (FOV), transcript density in FOV, and frequency of transcripts detected.

Generally, log10 transcript count > 4.0 in most area can be considered as a good quality standard for human tissue.

Need to note that the low accuracy in DAPI cell boundary is not a concern, as a self-designed cell segmentation processing will take over this task.

1.1.2.1 TT29258(Tumor)

1.1.2.2 NT29259(Normal)

1.1.3 Cancer-related Transcript Visaulization

• Blue: VEGFA (Fibroblast)
• MZB1: Yellow (B cell)
• Pink: SOX9 (Epithelial/Cancer cell)

Tumor Zoom-out

Tumor Zoom-in

Normal Zoom-out

Normal Zoom-in

2.1 Cell Segmentation & Filtering

Based on the spatial information and images obtained from MERFISH, we developed a machine learning model using the Cellpose algorithm to distinguish individual cells via MERFISH DAPI images.

To ensure the data quality and accuracy of cells, we have defined the minimum and maximum values for cell volume and gene count per cell. The cell volume should be between [100, 2000], and the gene count per cell > 25.

After that, we successfully obtained high-quality and accurate cell and gene data. We are now ready to perform statistical analysis.

For analysis platform, all of the analysis and visualization are performed using Scanpy, a commonly used Python package toolkit for analyzing single-cell gene expression data.

2.1.1 Transcript Count Violin

The transcript level is a basic aspect of data quality control. After filtering, we have ensured that the minimum transcript count per cell > 25.

Transcript Count Violin After Filtering

2.2 Dimension Reduction

We conducted Leiden UMAP clustering, a non-linear, manifold-aware dimension reduction algorithm, on our dataset to subset the cells into clusters. The following annotations are based on the umap clustering.

Umap and subclusters on all batch data

Umap colored by groups

3.1 Markers for type annotation

Cell markers dot plot, colored by expression

3.2 Cell Type Umap

Umap colored by cell type annotation

Umap colored by cell type annotation, split by group

3.3 Cell Type Spatial Map

With the annotations, we can map cell types back to their physical positions and create a spatial map for each sample.

Cell type spatial map, Tumor tissue

Cell type spatial map, Normal tissue

3.3 Higher Resolution Cell Subtypes

3.3.1 Epithelial cell

Epithelial cells Umap

Epithelial cells subtype markers umap

3.3.2 Fibroblast cell

Fibroblast Umap

Fibroblast subtype markers umap

Fibroblast subtype spatial map on Tumor Tissue

Fibroblast subtype spatial map on Tumor Tissue, Zoom-in

Fibroblast subtype spatial map on Normal Tissue

Fibroblast subtype spatial map on Normal Tissue, Zoom-in

3.4 Cell Type Qualification Analysis

Cell Count Barplot Comparision

4.1 all

4.2 Endothelial cell

4.3 Fibroblast

4.4 Lymphatic Endothelial cell

4.5 Macrophage

4.6 Mast cell

4.7 T cell

4.8 Inflammatory fibroblast

4.9 Myofibroblast

4.10 Tumor-like fibroblast

4.11 Vessel-associated fibroblast

4.12 inflame